e*****r
发帖数: 621 | 1 就是 mlogit model run 出来以后，要用一些 hypothetical independent variable
value 来算 predicted probability。请问在Stata 12下面具体的code是什么？
我知道在以前版本的Stata下，可以用 prvalue 来算，比如
prvalue, x(a=xx b=xx c=xx) rest(mean)
可现在发现 prvalue 不被recognize了。
另外知道在 LOGIT model里面，可以用 adjust a=xx b=xx c=xx, pr 但是发现这个在
MLOGIT 下面不行。
谢谢指教！ |
|
a******n
发帖数: 11246 | 2 我在build一个model的时候，用proc MI来impute missing values，然后得到一个
model equation。
现在我需要用这个equation来predict，那么input里的missing value怎么处理呢？用
同一个变量在build sample里的mean/median？还是有更好的办法？
Thanks! |
|
r********n
发帖数: 6979 | 3 build model的时候怎么impute的
predict的时候就怎么impute |
|
h***x
发帖数: 586 | 4 I mean, if we have campaign data, it should be easy to build a logistic
model to predict the probability to buy, but what if we do not have campaign
data?
I have a customer transaction data (only record the last purchase
behaviors, of course we have 3 party demographic data as well). The client
ask if we can provide the probability of customers' next purchase.Is
it possible? Any suggestion is welcome!
Thanks! |
|
a*****4
发帖数: 986 | 5 我的工作经历不长，先后做过这两种类型的marketing model。目前主要做后者，做多
了就感觉都是重复工作了，所以觉得ＭＭＭ更有挑战更有意思，也需要更多得business
的知识。大概因为predictive models发展更成熟了，但后者还在发展和推广之中，还
有很多争议。做ＭＭＭ的公司就是行业内的几家Ａd Ａgency，而后者有的实业公司也
有自己的部门做，是不是就业范围相对广些？
比较利弊，不知道哪块更有潜力，不知道是否要重新做回ＭＭＭ，矛盾中。
版上如果有同行，请分享你的看法。也欢迎讨论 |
|
l******n
发帖数: 9344 | 6 predictive modeling >> marketing modeling >> logistic regression
business |
|
n***p
发帖数: 508 | 7 Have anyone used time series independent variables to predict a binary
outcome? what methodology are you using.
for example, I have 12 month time series for each independent variable, and
at month 13, I have a binary outcome.
I want to keep the time series, and I do not want to transform the variables
like in the way people do general credit score card development.
please suggest, thanks |
|
z*****u
发帖数: 3010 | 8 最近面试 碰到一个问prediction for congressional elections。
感觉没有什么思路？只能随便忽悠过去了
有没有高手能够讲解？ 或者大家讨论一下
提点意见？ |
|
l******n
发帖数: 9344 | 9 no-buy client has 0 expected expenditure. If you have purchase quantity and
unit price, you can just predict the total expenditure as the product of
quantity and unit price.
that
working |
|
y**3
发帖数: 267 | 10 Thanks for the answer. It is very helpful. Will dig mroe aobut those two
models.
Can you elaborate a little for me?
So the models can be used to generate future predictions such as hazard
probabilities for the next month from date of censoring for scoring? Then
what assumption is used? Have you ever to put repeated event survival model
into production tp monitor fututre event?
Btw,I read somewhere the counting process is too easy to look into future.IS
that correct:) |
|
C******n
发帖数: 284 | 11 As far as I understand, you have to estimate a parametric model to predict
future survival time. |
|
y**3
发帖数: 267 | 12 thanks! Danou!
I did some research on repeated events survial model. As you said, a shared
frailty model fits my purpose well.
I have a question for what you said-For frailty models, could plug in the
empirical Bayes estimate of the prediction. Does this mean for time varying(
dependnet) covariates, that I dont need forecast their future path when
scoring to the near future?
thanks again
time. |
|
A*******s
发帖数: 3942 | 13 frailty and time dependent are not mutually exclusive. You could include
both in one model. As long as there is time dependent variable, you need to
forecast its path. Keep in mind the time dependent variable has to be
external and predictable.
shared
varying( |
|
s********0
发帖数: 2625 | 14 基本上没见过predictive model里放interaction的 |
|
a*****4
发帖数: 986 | 15 This model requests to predict a service provider preference - will assign
each individual with one of the 4 or 5 values which are different brands.
Data given: a survey responder file and a consumer extract file.
Has anyone had experience of this kind of modelling project? It would be
greatly appreciated that you can share some info/comments like modeling
method/process.....
I tried to build 5 1/0 logistic regression models and then score the
consumer extract file. However, how to weight the sc... 阅读全帖 |
|
|
N****s
发帖数: 645 | 17 请问哪位有电子版的 Predictive Modeling Using Logistic Regression Course
Notes - SAS 。
十分感谢。[email protected]/* */ |
|
N****s
发帖数: 645 | 18 请问哪位有电子版的 Predictive Modeling Using Logistic Regression Course
Notes - SAS 。
十分感谢。[email protected]/* */ |
|
P*****6
发帖数: 273 | 19 各位大拿：
有人知道为什么验证survival predictive model 用c- index 比较多。有什么更好的
方法吗？或者验证出发点不同的也可以。感觉光c-index不能满足需要。我很想知道单
个sample的预测，和实际的真实差距。当然因为有censored data, 直接比时间可能不
行。
多谢！ |
|
E*******r
发帖数: 2723 | 20 [申请ID]:
ElliottJr
[英文版名]:
Prediction
[中文版名]:
传统预测
[所属讨论区]:
娱乐休闲
[申请理由]:
八字推命、八卦、梅花易数、测字、五行生克理论等传统预测法有广泛基础，应该发扬
广大。
[治版纲领]:
不作恶。
[是否申请版主]:
否 |
|
l******n
发帖数: 9344 | 21 It is hard to predict income with these questions. I mean I try to ask as
many as possible. Now I include many projective questions. |
|
l******n
发帖数: 9344 | 22 I am trying predict income 10 years later, not now. |
|
g*****y
发帖数: 1120 | 23 click prediction水很深，很多时候ML做到后面就成了政治艺术。 |
|
w*****e
发帖数: 748 | 24 广告点击数据prior差不多是1/1000 到 1/10000. 看着数据不少，大部分variable都是
没用的。这么多variable还会有好多是假的，比如有一个人在north dakota 某个小村
点了一下，点击率100%，也得排除。
最要紧处理数据imbalance的问题。
decision tree基本上顶多1-2层，boosting好点，业界多用naive bayes。
另外，涉及到商业问题，要看click prediction干什么用，是为了提高点击率？还是为
了定价？提高点击率可以选没用的variable 去掉就行了，定价的话要选variance 最小
的就行了。
of |
|
m******e
发帖数: 21 | 25 CTR CVR Prediction业内应该可能有接近100篇相关的论文了 |
|
A*****n
发帖数: 243 | 26 seems very interesting
predictions
offense
it' |
|
z*****u
发帖数: 3010 | 27 最近面试 碰到一个问prediction for congressional elections。
感觉没有什么思路？
有没有高手能够讲解？ 或者大家讨论一下？
不提供数据，需要自己找
说是可参考下面这个网址 http://www.fec.gov/about.shtml |
|
b********r
发帖数: 37 | 28 【 以下文字转载自 JobHunting 讨论区 】
发信人: badweather (bad weather), 信区: JobHunting
标 题: 一个面试题（predictive model）
发信站: BBS 未名空间站 (Thu Mar 19 00:18:31 2015, 美东)
一家公司的面试题目：一个数据表格，只有两列，一列是表示实际天气情况（下雨-0，
不下雨-1），另一列是表示预测天气情况。表格一共365行，每行代表一天。从表格里
面我们可以得到error rate(预测错误的天数除以365)。现在问，有几种不同的方法预
测天气，我们能得到不同的error rate,问是不是error rate最低的方法最好？
我的回答是不一定：
1。error rate最低的有可能overfitting
2.这只是trainning data的fitting,如果用于实际预测，所选择的方法不一定最好，我
们要看看在test data上的performance.
面试官好像不是特别满意我的答案，而且告诉我没有另外的data,这怎么回答？ |
|
s****h
发帖数: 3979 | 29 昨天下午看到,觉得这是个很有趣,也实用的题目.
https://www.kaggle.com/c/restaurant-revenue-prediction
training data size: 137
test data size: 100K
Parameters: date, city name, city group, restaurant type, 37 number columns
for demographic , real estate, and commercial data.
初步设想, 用最基本的logistic regression?
city group 两个值, 那就是0, 1
restaurant type 四个值, 那就是0001, 0010,0100,1000
date可以换算成个数值
主要问题:
1. 参数这么多,training size这么小,一般有啥办法解决overfitting?
R2之类的步骤就没必要了吧.
2. 37个numeric column咋弄比较好? 估计有部分column就是category的标识,数值大小
没意义; ... 阅读全帖 |
|
s****h
发帖数: 3979 | 30 以前没怎么太follow过这些，刚看了一下http://www.kaggle.com/c/mlsp-2014-mri
第一名说他根本没有进一步改进model，因为在public LB上他排名很差。
估计第二名的情况也一样。
仔细一看，
最终第一（private LB)的在public LB上才排269
最终第一（private LB)的在public LB上才排282
而且搞笑的是他们的private score比public score高很多。？？？？
无数人overfitting啊。
用这些model裸run实在是有点搞笑啊！纯赌博啊！
更搞笑的是，这个restaurant-revenue-prediction的题目，很多人就猜测数据是怎么
产生的，想back engineer data. |
|
t*********u
发帖数: 26311 | 31 有人用过IBM的Predictive analytics产品么？ |
|
c*****o
发帖数: 169 | 32 训练了一个stacked model，回归问题，metamodel是random forest. 请问有什么办法
计算出prediction interval?
多谢! |
|
|
B*****e
发帖数: 2413 | 34 how about tomorrow's prediction? |
|
w*****y
发帖数: 3900 | 35 my prediction
W 09/01 Ga Tech 3:30PM
W 09/08 at Penn State 6:00 PM
L 09/15 at Michigan 3:30 PM
W 09/22 Mich St 3:30 PM
W 09/29 at Purdue TBD
L 10/06 at UCLA 8:00 PM
W 10/13 BC 3:30 PM
L 10/20 USC 3:30 PM
W 11/03 Navy 2:30 PM
W 11/10 Air Force 2:30 PM
W 11/17 Duke 2:30 PM
W 11/24 at Stanford 3:30 PM |
|
G******f
发帖数: 16223 | 36 use the expected average outcome to predict the outcome mah |
|
B*****m
发帖数: 679 | 37 ft.... i think there are more sophisticated models that predict outcomes. |
|
o**2
发帖数: 168 | 38 再贴点运行结果，供没空动手的同学参考。
Predictor#0 is starting - main
Predictor#0 is ending
Predictor#0 predict(input0) is starting - main
Predictor#0 predict(input0) is ending
Predictor#0 predict(input1) is starting - main
Predictor#0 predict(input1) is ending
Predictor#0 predict(input2) is starting - main
Predictor#0 predict(input2) is ending
Predictor#0 predict(input3) is starting - main
Predictor#0 predict(input3) is ending
Predictor#0 predict(input4) is starting - main
Predictor#0 predict(input4) is ending
Pr... 阅读全帖 |
|
n*********8
发帖数: 1137 | 39 美国病毒改造监控计划始末：为改造全新病毒 从中国拿走上万件样本？
送交者: 两极的世界[♂★★★声望勋衔13★★★♂] 于 2020-02-29 17:41 已读 469
次 大字阅读 两极的世界的个人频道
“最危险的人造病毒”：病毒学家争相改造病毒研究引发公共安全担忧
2014年10月17日下午，病毒学家Ralph Baric博士没有照例来到他在北卡罗来纳大学教
堂山分校的实验室，他正在为自己女儿周末的婚礼做最后的准备。 此时的他并不知道
，奥巴马政府由于对实验室使用SARS, MERS和流感等病毒，进行功能获得性(gain of
function)改造的研究，对公共健康构成的潜在威胁的担忧，通过白宫科技政策办公室
联合美国卫生与公众服务部发布禁令，宣布中止对类似研究的资金资助[1]，并要求相
关领域的研究人员立即停止相关的研究，直到研究项目的风险和收益被美国国家生物安
全科学顾问委员会(NSABB)和美国国家学院国家科学研究委员会(NRC)的专家进行评估后
方可继续[2]。
这其中就有武汉病毒所石正丽博士的研究团队与Ralph Baric博士的合作项目：一个正
在进... 阅读全帖 |
|
l**i
发帖数: 8144 | 40 简单地讲 就是谋事在人 成事在天
认为社会发展沿着某个预定方向发展的理论 都是傻逼理论
--------------------------------------
Anthropic principle
From Wikipedia, the free encyclopedia
In astrophysics and cosmology, the anthropic principle is the philosophical
argument that observations of the physical Universe must be compatible with
the conscious life that observes it. Some proponents of the argument reason
that it explains why the Universe has the age and the fundamental physical
constants necessary to accommodate conscious life. As a ... 阅读全帖 |
|
f**d
发帖数: 768 | 41 这是一本计算神经科学的优秀著作，全文拷贝这里（图和公式缺），有兴趣的同学可以
阅读
如需要，我可以分享PDF文件（－－仅供个人学习，无商业用途）
From Computer to Brain
William W. Lytton
From Computer to Brain
Foundations of Computational Neuroscience
Springer
William W. Lytton, M.D.
Associate Professor, State University of New York, Downstato, Brooklyn, NY
Visiting Associate Professor, University of Wisconsin, Madison
Visiting Associate Professor, Polytechnic University, Brooklyn, NY
Staff Neurologist., Kings County Hospital, Brooklyn, NY
In From Computer to Brain: ... 阅读全帖 |
|
o**2
发帖数: 168 | 42 Give each Predictor object its own thread, and apply producer/consumer
pattern twice: one is from user thread to predictor thread for dropping off
prediction input, and the other one is form predictor thread to user thread
for picking up computed prediction.
public class Predictor implements Runnable {
private String input, prediction;
private boolean stop;
public Predictor () {
new Thread (this).start ();
}
public synchronized boolean isIdle () {
return input... 阅读全帖 |
|
l****z
发帖数: 29846 | 43 A new year is around the corner, and some climate scientists and
environmental activists say that means we're one step closer to a climate
Armageddon. But are we really?
Predicting the weather -- especially a decade or more in advance -- is
unbelievably challenging. What's the track record of those most worried
about global warming? Decades ago, what did prominent scientists think the
environment would be like in 2010? FoxNews.com has compiled eight of the
most egregiously mistaken predictions, ... 阅读全帖 |
|
l****z
发帖数: 29846 | 44 Eight Botched Environmental Forecasts
By Maxim Lott
Published December 30, 2010
FoxNews.com
A new year is around the corner, and some climate scientists and
environmental activists say that means we're one step closer to a climate
Armageddon. But are we really?
Predicting the weather -- especially a decade or more in advance -- is
unbelievably challenging. What's the track record of those most worried
about global warming? Decades ago, what did prominent scientists think the
environment would be... 阅读全帖 |
|
p********a
发帖数: 6437 | 45 honestly, I'd like to further discuss this with you so that your post is NOT
misleading the others.
u get rich, that's nice. but your conclusion is still NOT right.
1. 操作方法和资金管理的作用 is useless if you cannot predict the market
look, my logic is simple:
prediction is useful --> a trader will win
in your logic, if 操作方法和资金管理 is useful ALONE, why not u just
enter the market anytime using a coin-toss? or anytime if you wish?
so in the end of the day, u still need an edge (prediction) ... 阅读全帖 |
|
l****z
发帖数: 29846 | 46 http://dailycaller.com/2016/04/22/7-enviro-predictions-from-ear
Environmentalists truly believed and predicted during the first Earth Day in
1970 that the planet was doomed unless drastic actions were taken.
Humanity never quite got around to that drastic action, but
environmentalists still recall the first Earth Day fondly and hold many of
the predictions in high regard.
So this Earth Day, The Daily Caller News Foundation takes a look at
predictions made by environmentalists around the original... 阅读全帖 |
|
S*********4
发帖数: 5125 | 47 外国网民猜测中国何时赶超美国(转载)点击：8389 回复：41 作者：上帝使者笑了
发表日期：2010-12-24 19:05:00 回复
【译文-中文】【经济学人】中国来了！预测号
【原文-英国】save the date
【原文地址】：http://www.economist.com/blogs/d ... date?page=1&sort=recommend
【译文来源】：网贴翻译论坛
【翻译作者】：康康Tainy
【联名作者】：传输用
【译文地址】：http://www.ptfcn.com/ptfcn/j/140.html
Save the date
Dec 16th 2010, 12:31 by P.W. and D.H
We invite you to predict when China will overtake America
CHINA jumped ahead of Japan in 2010 to become the world’s second-
biggest economy, but when will it grab the nu... 阅读全帖 |
|
g*****1
发帖数: 666 | 48 常家蹲还是重复他的念咒一般的老生常谈，根本不值一提。
可读是读者的评论。
http://www.foreignpolicy.com/articles/2011/12/29/the_coming_col
MESS_MEDIA
9:18 PM ET
December 29, 2011
Yeah, I was wrong when I
Yeah, I was wrong when I predicted that the world would end on May 21, 2011.
It should end on October 21.
-Harold Camping
REPLY
JOESMITHII
2:44 AM ET
December 30, 2011
I guess, sooner or later,
I guess, sooner or later, Gordon will get used to admitting wrong
predictions. Fortunately, 2012 is around the corner, we don't need to wa... 阅读全帖 |
|
发帖数: 1 | 49 Predictions and Prophecies of Jeane Dixon
http://www.visiontimes.com/2017/10/04/predictions-and-prophecies-of-jeane-dixon.html
October 4, 2017
Jeane Dixon, a renowned American psychic, predicted the death of President
Franklin Roosevelt, the defeat of Prime Minister Winston Churchill in an
election, and the assassination of President John F. Kennedy. Dixon also
predicted that China would be taken over by communists. Before she passed
away in 1997, she predicted that the hope of humankind lay in ... 阅读全帖 |
|
R*****h
发帖数: 194 | 50 Consumer Reports' 2010 Annual Car Reliability Survey: GM Makes Big Strides
While Honda, Toyota Dominate
SEE ALSO: New Car Buyers Guide - Compare Specs, Prices, Total Ownership
Costs and Crash Ratings
European reliability rally stalls: Audi, BMW, and Mercedes-Benz among the
worst
YONKERS, N.Y., Oct. 26, 2010; General Motors has improved considerably,
though Honda and Toyota still dominate in the latest predicted-reliability
Ratings of new cars, according to Consumer Reports' 2010 Annual Auto Surv... 阅读全帖 |
|
